Related Experiment Video
Updated: Apr 23, 2026
05:58
Retinal Detachment Model in Rodents by Subretinal Injection of Sodium Hyaluronate
Published on: September 11, 2013
21.7K
RetinaDetachNet: Automated Deep Learning Quantification of Photoreceptor Cell Death for Neuroprotection Studies in
Konstantinos G Baroutis1, Hani El Helwe1, Kaho Yamamoto2
1Department of Ophthalmology, Retina Service, Ines and Frederick Yeatts Lab in Retina Research, Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA.
Translational Vision Science & Technology
|April 22, 2026
Summary
RetinaDetachNet, a novel deep learning tool, accurately quantifies TUNEL-positive cells in retinal detachment models. This validated pipeline accelerates research into photoreceptor cell death and potential neuroprotective therapies.
Area of Science:
- Ophthalmology
- Computational Biology
- Neuroscience
Background:
- Retinal detachment (RD) is a significant cause of vision loss.
- Quantifying photoreceptor cell death is crucial for understanding RD pathogenesis and testing therapies.
- Current methods for TUNEL-positive cell counting are often time-consuming and subjective.
Purpose of the Study:
- To develop and validate RetinaDetachNet, the first deep learning pipeline for automated quantification of TUNEL-positive cells in experimental retinal detachment models.
- To assess the accuracy and reproducibility of RetinaDetachNet compared to manual counting.
- To provide an open-source tool for observer-independent analysis of photoreceptor cell death.
Main Methods:
- RetinaDetachNet integrates a U-Net for outer nuclear layer (ONL) segmentation and a hybrid approach for TUNEL-positive cell detection using StarDist and Otsu thresholding.
- Validation was performed on three independent datasets (primary, historical, external) with varying imaging parameters.
- Agreement with manual counts was evaluated using Spearman correlation and Bland-Altman analysis.
Main Results:
- The U-Net achieved a high Dice coefficient of 0.93 for ONL segmentation.
- RetinaDetachNet demonstrated strong correlation with manual counts across all datasets (ρ = 0.80–0.98), outperforming individual StarDist or Otsu methods.
- Bland-Altman analysis confirmed minimal systematic bias, indicating high agreement.
Conclusions:
- RetinaDetachNet is the first validated deep learning pipeline for TUNEL-positive cell quantification in retinal detachment models.
- The tool offers superior accuracy and reproducibility through its hybrid dual-validation architecture.
- RetinaDetachNet significantly reduces analysis time, enabling faster evaluation of neuroprotective agents.
